Google's New AI Tool Predicts Flash Floods Up to 24 Hours in Advance
Google has developed Groundsource, an AI-powered system that addresses the lack of historical data necessary for accurately predicting flash floods in urban areas. Using Gemini AI, Groundsource scans millions of global news articles since 2000 to extract and geolocate mentions of flash floods, creating a dataset of 2.6 million events across over 150 countries. This dataset allows Google to train an LSTM neural network to forecast flash flood risks up to 24 hours in advance for urban areas, factoring in weather, urban density, soil absorption, and topography. The predictions are now available on Flood Hub, Google's existing flood warning platform. While the system can only assess areas of about 20 square kilometers, and its accuracy is limited by the availability of news coverage, it provides timely, actionable alerts that have already assisted emergency responses in the field. Groundsource effectively turns news reports into vital data, aiming to improve disaster preparedness and resilience against sudden urban flash flooding.
